July was the month we opened a new front. After years of building distributed infrastructure for AI workloads, we launched our AI gaming services vertical and shipped its first product, AI Characters, then followed it with a public developer repository so teams can wire AI-driven NPCs into a real game loop. Alongside that, the validator council grew again, our research team presented at one of the premier conferences in distributed systems, and we rebuilt the EdgeCloud homepage. Here’s everything that happened.
ZAN, Ant Digital Technologies’ Web3 Brand, Joins as a Strategic Enterprise Validator
ZAN, the Web3 brand of Ant Digital Technologies, joined the Theta ecosystem as a strategic Enterprise Validator Node operator, taking its place alongside Google, Samsung, Sony, CAA, Binance, Deutsche Telekom, Docomo Global and Cloudician. The node handles transaction validation, block production, network security and on-chain governance, staking THETA to secure the consensus layer our decentralized cloud runs on. As a subsidiary of Ant Group, ZAN brings compliance depth and real-world asset experience from one of Asia’s largest payments and technology platforms.
Theta Labs Launches AI Gaming Services and AI Characters
We launched our AI gaming services vertical, and AI Characters is the first product built for it. Developers can define a game world, create characters with personalities, speaking styles, goals and knowledge boundaries, then run sessions in parallel, with characters returning machine-readable decisions that feed straight into a game loop rather than a chat window. You can try it in the poker demo at ai-game-demo.thetaedgecloud.com, or start building at thetaedgecloud.com/dashboard/gaming/characters.
The AI Characters Developer Repository Goes Public
We followed the launch by publishing the AI Characters developer repository, which takes the product from concept to something a team can run locally the same day. It covers defining characters, placing several in shared sessions, feeding them game state and events, and the part that makes this usable inside a real game rather than a chat window: requesting machine-readable decisions that feed directly into a game loop. Inside are a working reference implementation, setup instructions, provisioning scripts, the Characters API specification, and an Agent Skill for Claude Code, Cursor and other compatible tools. The included project uses poker, but the pattern adapts to other genres and gameplay systems.
Repository: https://github.com/thetatoken/theta-edge-cloud-ai-characters-public
Build your NPC dialogue: https://www.thetaedgecloud.com/dashboard/gaming/characters
ShuttleCross Presented at IEEE/IFIP DSN 2026
We were glad to see joint research between Theta and Professor Zhen Xiao’s team presented at IEEE/IFIP DSN 2026 in Charlotte this month, one of the premier conferences in distributed systems and reliability. The paper introduces ShuttleCross, a framework for cross-chain smart contract invocation that guarantees atomicity and serializability while addressing the latency, aborts and limited concurrency that existing protocols suffer from.
Read the paper here: https://blog.thetatoken.org/content/files/zhenxiao-com/papers/shuttlecross.pdf
A Rebuilt EdgeCloud Homepage
We rebuilt the EdgeCloud homepage from the ground up. GPU compute, inference endpoints, AI agents and the developer toolkit now sit together in one place, so it is clear at a glance what you can run and what it costs. As the product surface has grown, so has the job of explaining it, and a lot of work went into getting this one right.
See it at thetaedgecloud.com.
Theta EdgeCloud Powers Human-Centered AI Research at Soongsil University
Soongsil University’s HUMANE Lab became the 36th member of our academic network. Led by Professor Kunwoo Park within the College of AI, the lab studies language and multimodal AI built on large language models, focused on societal problems such as misinformation, bias and fairness. Their teams took first place at the FEVER 2026 challenge (EACL) after back-to-back second place finishes in 2025 and 2024, and their winning system verifies image-text claims in four stages, coordinating multiple language and embedding models in parallel, exactly the kind of workload EdgeCloud’s 30,000+ node network is built to serve.
